Defect detection management model for automated optical inspection equipment
By generating interpretable, reusable, and auditable defect detection execution rules through structured analysis of acceptance requirements, the problem of automatic optical inspection equipment being sensitive to process and material fluctuations is solved. This enables stable defect detection and low false defect probability in fluctuating environments, thereby reducing maintenance costs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHONGJIA MICROVISION (SHENZHEN) SEMICONDUCTOR TECHNOLOGY CO LTD
- Filing Date
- 2026-03-19
- Publication Date
- 2026-07-17
AI Technical Summary
Existing automatic optical inspection equipment defect detection models are sensitive to process and material fluctuations, have high maintenance costs, rely on a large amount of labeled data and lack interpretability, making it difficult to achieve cross-batch transfer. Furthermore, on-site acceptance criteria and handling actions are expressed in natural language, making it difficult to directly translate them into executable defect detection strategies.
A defect detection management model for an automated optical inspection device is provided, including a requirement access module, a constraint parsing module, a threshold-related parameter generation module, a rule writing module, and a statistics and calibration module. By structuredly parsing acceptance requirements, it generates interpretable, reusable, and auditable defect detection execution rules and maintains stable defect detection capabilities under fluctuating process environments.
It achieves stable defect detection capability under fluctuating process conditions, reduces the probability of false defects, improves the generalization and robustness of detection, reduces maintenance costs, and provides an interpretable and reusable detection and analysis system.
Smart Images
Figure CN122415446A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of defect detection technology, and specifically to a defect detection management model for an automated optical inspection device. Background Technology
[0002] In the manufacturing process of display panels and semiconductors, automated optical inspection equipment is typically used to automatically detect defects in the inspected devices. However, the rule / template methods in the defect detection models of these automated optical inspection equipment are sensitive to fluctuations in processes, materials, and structures, resulting in high maintenance costs. Furthermore, the deep learning methods used in these defect detection models rely on large amounts of labeled data, making cross-batch transfer difficult and lacking in interpretability. Finally, on-site acceptance criteria and handling actions are often expressed in natural language, making it difficult to directly translate them into executable defect detection strategies and reusable defect detection formulas. Summary of the Invention
[0003] The main objective of this disclosure is to provide a defect detection management model for an automated optical inspection device to solve at least some of the technical problems shown above.
[0004] To achieve the above objectives, the first aspect of this disclosure provides a defect detection management model for an automated optical inspection device, the defect detection management model comprising: The demand access module is used to receive acceptance requirements for the device under test. The constraint parsing module is used to perform structured parsing of the acceptance requirements to obtain at least one structured constraint indicator. A threshold-related parameter generation module is used to generate the threshold parameters for each of the structured constraint indicators; The rule writing module is used to generate defect detection execution rules based on the structured constraint indicators and the indicator threshold parameters of each structured constraint indicator; The statistics and calibration module is used to adjust the index threshold parameters based on the actual defects of the device under test and the detected defects obtained by executing the defect detection execution rules, and to update the defect detection execution rules based on the adjusted index threshold parameters.
[0005] In some embodiments of this disclosure, the structured constraint indicators include at least some of the following indicators: detection area indicator of the device under test, defect type indicator, false detection risk budget indicator, acceptance standard indicator, allowable threshold range indicator, exception case indicator, defect detection execution action indicator, defect interpretation indicator, acceptance consistency indicator, maintenance cost indicator, and defect detection coverage indicator.
[0006] In some embodiments of this disclosure, the rule writing module determines the defect detection execution rule for each structured constraint indicator and its corresponding indicator threshold parameter, based on the large language model, specific domain language model, rule template library and parameter binding rules corresponding to the structured constraint indicator. The defect detection management model also includes a recipe encapsulation module, which is used to encapsulate the defect detection execution rules of all the structured constraint indicators based on versioned encapsulation rules to obtain the defect detection execution recipe version. The defect detection execution recipe version also includes syntax verification execution rules, range verification execution rules, evidence field constraint execution rules, gray release execution rules, and rollback point execution rules.
[0007] In some embodiments of this disclosure, the statistics and calibration module includes: The actual defect statistics module is used to collect actual defect data of the tested device; The defect detection statistics module is used to statistically analyze the defect detection data obtained by executing the defect detection execution rules. The verification guardrail module is used to calculate the actual false detection data from the actual defect data and the detected defect data, and to determine whether the actual false detection data meets the false detection risk budget index in the structured constraint index. The quantile mapping module is used to process the actual defect data and the detected defect data to obtain the quantile mapping result; The threshold adaptive module is used to generate suggested threshold parameters for the index based on the actual false detection data and the quantile mapping result when the actual false detection data does not meet the false detection risk budget index in the structured constraint index.
[0008] In some embodiments of this disclosure, the threshold adaptive module is further configured to: When the actual false detection data does not meet the false detection risk budget index in the structured constraint index, a change description and a change impact assessment are generated based on the actual false detection data and the quantile mapping result.
[0009] In some embodiments of this disclosure, the defect detection management model further includes: a version and canary release module, which is used to update the suggested parameter of the indicator threshold to the indicator threshold parameter based on the canary release execution rule; The threshold adaptive module is further configured to: after updating the indicator threshold suggestion parameter to the indicator threshold parameter, store the defect detection execution recipe version before the change of the indicator threshold parameter and the defect detection execution recipe version after the change to obtain a recipe version record, which can be used when rolling back according to the rollback point execution rule in the defect detection execution rule.
[0010] In some embodiments of this disclosure, the defect detection management model further includes: An audit and replay module is used to monitor KPI trends when executing the defect detection recipe version; The KPI trends include: acceptance consistency trend, maintenance cost trend, and formula defect detection coverage trend, which change over time as the defect detection is performed on the formula version.
[0011] In some embodiments of this disclosure, the defect detection management model further includes: The scoring module is used to score the currently executed defect detection execution recipe version based on the KPI trend during the execution of the defect detection execution recipe version; The rollback module, wherein the rollback point execution rules include: when the score of the currently executed defect detection execution recipe version is less than the score of the historically stored defect detection execution recipe version in the recipe version record, the rollback module rolls back the defect detection execution recipe version to the historically stored defect detection execution recipe version based on the recipe version record, so that the score of the currently executed defect detection execution recipe version is the highest-scoring defect detection execution recipe version in the recipe version record.
[0012] In some embodiments of this disclosure, the scoring module obtains a score for the currently executed defect detection execution recipe version based on the KPI trend, including: The scoring module obtains an acceptance consistency score based on the acceptance consistency trend. The scoring module generates a maintenance cost score based on the maintenance cost trend. The scoring module obtains a formula defect detection coverage score based on the trend of the formula defect detection coverage. The scoring module calculates the score of the currently executed defect detection execution recipe version by weighting the acceptance consistency score, the maintenance cost score, and the recipe defect detection coverage score.
[0013] In some embodiments of this disclosure, the scoring module performs a weighted calculation based on the acceptance consistency score, the maintenance cost score, and the recipe defect detection coverage score to obtain a score for the currently executed defect detection execution recipe version, including: The scoring module generates acceptance consistency weight, maintenance cost weight, and defect detection coverage weight based on the acceptance consistency index, maintenance cost index, and defect detection coverage index in the structured constraint indicators, respectively. The scoring module multiplies the acceptance consistency score by the acceptance consistency weight to obtain a first product; The scoring module multiplies the maintenance cost score by the maintenance cost weight to obtain a second product; The scoring module multiplies the formula defect detection coverage score by the defect detection coverage weight to obtain a third product; The scoring module uses the sum of the first product, the second product, and the third product as the score for the currently executed defect detection execution recipe version.
[0014] The second aspect of this disclosure provides an automated optical inspection device, which includes a defect detection management model of any of the automated optical inspection devices provided in the first aspect of this disclosure.
[0015] The defect detection management model of the automated optical inspection equipment provided in this disclosure uses a constraint parsing module to perform structured parsing of acceptance requirements to obtain at least one structured constraint index. A threshold-related parameter generation module generates index threshold parameters for each structured constraint index. A rule writing module generates defect detection execution rules based on the structured constraint index and the index threshold parameters for each structured constraint index. A statistics and calibration module adjusts the index threshold parameters based on the actual defects of the device under test and the detected defects obtained by executing the defect detection execution rules, and updates the defect detection execution rules based on the adjusted index threshold parameters. This provides an interpretable, reusable, and auditable detection and analysis system that maintains stable defect detection capability and reduces the probability of false defects under fluctuating operating conditions. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the specific embodiments or related technologies of this disclosure, the accompanying drawings used in the description of the specific embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1A schematic block diagram of a defect detection management model for an automated optical inspection device provided in an embodiment of this disclosure; Figure 2 A flowchart illustrating the operation of a defect detection management model for an automated optical inspection device provided in an embodiment of this disclosure; Figure 3 A schematic block diagram of a defect detection management model for an automated optical inspection device provided in another embodiment of this disclosure; Figure 4 A schematic block diagram of a statistics and calibration module provided in an embodiment of this disclosure; Figure 5 A schematic block diagram of a defect detection management model for an automated optical inspection device provided in another embodiment of this disclosure; Figure 6 A flowchart illustrating the overall process of requirements understanding → threshold adaptation → rule / recipe output provided for one embodiment of this disclosure; Figure 7 A schematic diagram of a structured constraint index provided in an embodiment of this disclosure; Figure 8 A schematic flowchart illustrating the rule writing and recipe packaging process provided in one embodiment of this disclosure; Figure 9 A flowchart illustrating the threshold adaptation process provided in one embodiment of this disclosure; Figure 10 This is a schematic diagram illustrating a KPI trend according to an embodiment of the present disclosure; Figure 11 This is a schematic diagram illustrating tabular output according to an embodiment of the present disclosure; Figure 12 This is a schematic diagram illustrating a deployment architecture according to an embodiment of the present disclosure; Figure 13 This is a comparison diagram showing the effects of the model illustrated in the embodiments of this disclosure and existing methods. Detailed Implementation
[0018] To enable those skilled in the art to better understand the present disclosure, the technical solutions of the present disclosure will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present disclosure, and not all embodiments. Based on the embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present disclosure.
[0019] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this disclosure described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0020] In this disclosure, the terms “upper,” “lower,” “left,” “right,” “front,” “rear,” “top,” “bottom,” “inner,” “outer,” and “middle,” etc., indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings. These terms are primarily for the purpose of better describing this disclosure and its embodiments, and are not intended to limit the indicated devices, elements, or components to having a specific orientation, or to be constructed and operated in a specific orientation.
[0021] Furthermore, in addition to indicating location or positional relationship, some of the aforementioned terms may also have other meanings. For example, the term "above" may also be used in certain circumstances to indicate a dependency or connection. Those skilled in the art can understand the specific meaning of these terms in this disclosure according to the specific circumstances.
[0022] Furthermore, the terms "set up," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral structure; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection via an intermediate medium, or an internal connection between two devices, components, or parts. Those skilled in the art can understand the specific meaning of these terms in this disclosure according to the specific circumstances.
[0023] It should be noted that, unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other. This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.
[0024] Example 1 To address at least some of the aforementioned problems, embodiments of this disclosure provide a defect detection management model for an automated optical inspection device. This model relates to the fields of industrial automated optical inspection (AOI) and intelligent analysis, and particularly to defect distribution modeling, feature space analysis, semantic reasoning, and strategy output for the inspected device. The inspected device can be, but is not limited to, display panels, wafers, etc.
[0025] refer to Figure 1 , Figure 2 and Figure 6 The defect detection management model mainly includes: a requirement access module, a constraint parsing module, a threshold-related parameter generation module, a rule writing module, and a statistics and calibration module. The requirement access module receives acceptance requirements for the device under test, which are expressed in natural language as acceptance criteria, risk preferences, and corresponding actions. For example, these acceptance requirements can be natural language requirements, acceptance criteria, risk preferences, and acceptance-related statistical data.
[0026] refer to Figure 1 , Figure 2 and Figure 6 The constraint parsing module is used to perform structured parsing of acceptance requirements to obtain at least one structured constraint indicator. In other words, the constraint parsing module parses the acceptance requirements to obtain structured constraint indicators, which serve as the entry point for generating defect detection execution rules. The threshold-related parameter generation module is used to generate the threshold parameters for each structured constraint indicator.
[0027] refer to Figure 1 , Figure 2 and Figure 6 The rule writing module is used to generate defect detection execution rules based on structured constraint indicators and the indicator threshold parameters for each structured constraint indicator. For example, defect detection execution rules can be automatically generated based on DSL / configuration.
[0028] refer to Figure 1 , Figure 2 and Figure 6 The statistics and calibration module is used to adjust the index threshold parameters based on the actual defects of the device under test and the detected defects obtained by executing the defect detection execution rules, and to update the defect detection execution rules based on the adjusted index threshold parameters.
[0029] In the above scheme, the defect detection management model of the provided automated optical inspection equipment uses a constraint parsing module to perform structured parsing of acceptance requirements to obtain at least one structured constraint index. A threshold-related parameter generation module generates threshold parameters for each structured constraint index. A rule writing module generates defect detection execution rules based on the structured constraint index and its threshold parameters. A statistics and calibration module adjusts the threshold parameters based on the actual defects of the inspected device and the detected defects obtained by executing the defect detection execution rules, and updates the defect detection execution rules based on the adjusted threshold parameters. (Reference) Figure 13Therefore, the beneficial effects of the solution provided in this disclosure compared to related solutions are: it provides an interpretable, reusable, and auditable detection and analysis system that maintains stable defect detection capabilities and reduces the probability of false defects under fluctuating operating conditions. The defect detection management model of the automated optical inspection equipment provided in this disclosure can maintain stable judgment criteria under fluctuating process / structure environments, improving generalization and robustness.
[0030] The following is in conjunction with the appendix Figures 1 to 13 The defect detection management model of the automated optical inspection equipment disclosed herein is described in detail.
[0031] The types of structured constraint indicators can include various methods, some of which are illustrated below. For example, refer to... Figure 7 and Figure 11 Structured constraint indicators can include at least some of the following: detection area indicators, defect type indicators, false positive risk budget indicators, acceptance standard indicators, allowable threshold range indicators, exception case indicators, defect detection execution action indicators, defect interpretation indicators, acceptance consistency indicators, maintenance cost indicators, and defect detection coverage indicators. By including exception case indicators in the structured constraint indicators, the probability of false defects detected by automated optical inspection equipment can be reduced, thereby improving defect detection accuracy. By including false positive risk budget indicators in the structured constraint indicators, the false positive risk of automated optical inspection equipment can be controlled to meet acceptance requirements. By incorporating defect interpretation indicators into the structured constraint indicators, the defect detection management model of automated optical inspection equipment can output an interpretable detection and analysis system. By incorporating acceptance consistency indicators into the structured constraint indicators, the detection capability of the defect detection management model can meet acceptance consistency requirements. By incorporating maintenance cost indicators into the structured constraint indicators, the maintenance cost of the defect detection management model can meet preset acceptance requirements. By incorporating the defect detection coverage index into the structured constraint index, the defect detection coverage of the defect detection management model can be made to meet the preset acceptance requirements.
[0032] There are several ways to generate defect detection execution rules in the rule writing module. Some examples are described below. For example, refer to... Figure 8The rule writing module, for each structured constraint indicator and its corresponding threshold parameters, determines the defect detection execution rules for each structured constraint indicator based on the large language model, domain-specific language model, rule template library, and parameter binding rules corresponding to that structured constraint indicator. That is, the defect detection management model disclosed in this disclosure is not for a single type of defect detection model, but rather a management model for most or even all defect detection models running on automated optical inspection equipment. The rule writing module uses the threshold parameters of each structured constraint indicator to call, adjust, and even manage the defect detection status of the defect detection model corresponding to that structured constraint indicator, so that the defect detection capability of the defect detection model corresponding to that structured constraint indicator meets the structured constraint indicator, thereby satisfying acceptance requirements.
[0033] For example, refer to Figure 3 and Figure 6 The defect detection management model also includes a recipe encapsulation module. This module encapsulates the defect detection execution rules for all structured constraint indicators based on versioned encapsulation rules, resulting in defect detection execution recipe versions. Specifically, the recipe encapsulation module can select some or all of the structured constraint indicator defect detection execution rules from the entire set of structured constraint indicator defect detection execution rules and perform combined execution in a permutation and combination manner, thereby obtaining multiple different types of defect detection execution recipe versions. At any given time, the automated optical inspection equipment can only execute one defect detection execution recipe version, thus simultaneously executing one or more structured constraint indicator defect detection execution rules contained within that recipe version. This allows for the merging of non-interfering defect detection indicators to improve defect detection speed, while also facilitating the separate detection of interfering defect detection indicators to improve defect detection accuracy.
[0034] For example, refer to Figure 8The defect detection execution recipe version can also include syntax validation execution rules, range validation execution rules, evidence field constraint execution rules, gray-scale release execution rules, and rollback point execution rules. For example, the syntax validation execution rules and evidence field constraint execution rules can perform legality checks (fieldization / evidence constraints). For example, syntax, range, and evidence constraint checks can be performed on the defect detection execution rules or defect detection execution recipe version; failures result in downgrading or blocking. Through this method, during the recipe encapsulation process of defect detection execution rules for all structured constraint indicators, a second syntax check can be performed to identify errors or mutual interference in the defect detection model. Furthermore, evidence field constraint execution rules, gray-scale release execution rules, and rollback point execution rules can be provided, facilitating subsequent evaluation of the execution effect of the defect detection execution recipe version and determining whether to roll back to a previous defect detection execution recipe version, thus ensuring the defect detection effect of the automated optical inspection equipment.
[0035] There are several ways to configure the statistics and calibration module, some of which are illustrated below.
[0036] For example, refer to Figure 4 and Figure 9 The statistics and calibration module includes an actual defect statistics module, a detected defect statistics module, a verification guardrail module, a quantile mapping module, and a threshold adaptation module. The actual defect statistics module is used to collect actual defect data of the device under test. The detected defect statistics module is used to collect detected defect data obtained by executing defect detection rules. The verification guardrail module is used to calculate actual false detection data from the actual defect data and detected defect data, and to determine whether the actual false detection data meets the false detection risk budget index in the structured constraint index. The quantile mapping module is used to process the actual defect data and detected defect data to obtain quantile mapping results. The threshold adaptation module is used to generate suggested threshold parameters based on the actual false detection data and quantile mapping results when the actual false detection data does not meet the false detection risk budget index in the structured constraint index. The actual defect data can be used as a reference baseline, and the detected defect data can represent the current defect detection situation. The threshold adaptation module generates suggested threshold parameters based on the reference baseline, the current defect detection situation distribution, and the false detection risk budget index. In this way, during the defect detection process of the automated optical inspection equipment, the false detection risk budget index is detected in real time. When the false detection risk budget index does not meet the acceptance requirements, the threshold adaptive module generates suggested index threshold parameters based on the actual false detection data and quantile mapping results, thereby providing suggestions for optimizing the index threshold parameters so that the automated optical inspection equipment meets the false detection risk budget index.
[0037] For example, refer to Figure 9 The threshold adaptive module is also used to generate change descriptions and change impact assessments based on the actual false detection data and quantile mapping results when the actual false detection data does not meet the false detection risk budget indicators in the structured constraint indicators. In this embodiment, the threshold adaptive module not only provides suggested parameters for indicator thresholds, but also provides change descriptions and change impact assessments, making it easier for staff or defect detection management models to accurately assess the impact of changes to indicator threshold parameters by combining the change descriptions and change impact assessments, thereby improving the ability to optimize defect detection execution recipe versions. For example, the above-mentioned change descriptions and change impact assessments can be: field-based explanations and change impact assessment records output based on the above-mentioned defect interpretation indicator rules, evidence field constraint execution rules, and syntax verification execution rules.
[0038] For example, refer to Figure 5 The defect detection management model may also include a version and canary release module. This module updates the suggested parameter parameters of the indicator thresholds to the actual indicator threshold parameters based on the canary release execution rules. By updating the suggested parameter parameters to the actual indicator threshold parameters based on the canary release execution rules, the version and canary release of defect detection execution recipes, as well as the rollback of indicator threshold parameters, are based on drift-triggered defect detection. This avoids the uncontrollable or difficult-to-control impact of sudden, one-time adjustments to the suggested parameter parameters, achieving gradual optimization of the indicator threshold parameters and improving the consistency of defect detection result output.
[0039] For example, refer to Figure 9The threshold adaptive module can also be used to: after updating the suggested threshold parameter to the actual threshold parameter, store the defect detection execution recipe version before the change and the changed defect detection execution recipe version to obtain a recipe version record, which can be used when rolling back according to the rollback point execution rule in the defect detection execution rule. That is, after the indicator threshold parameter is optimized (the indicator threshold parameter changes from the current value to the suggested indicator threshold parameter), the rule writing module will regenerate the defect detection execution rule based on the latest indicator threshold parameter corresponding to the structured constraint indicator; the recipe encapsulation module will re-encapsulate the recipe based on the updated defect detection execution rule to obtain the latest version of the defect detection execution recipe version. At this time, the threshold adaptive module does not delete the defect detection execution recipe version before the change of the indicator threshold parameter, but retains and stores it for use when rolling back according to the rollback point execution rule, thereby improving the reusability of the defect detection execution recipe version, realizing a reusable detection and analysis system, so as to maintain stable defect detection capability and reduce the probability of false defects under fluctuating operating conditions. In the above approach, the defect detection execution recipe version can be accumulated into a reusable defect detection execution recipe version asset, which can be reused across scenarios and reduce long-term maintenance costs.
[0040] For example, refer to Figure 5 The defect detection management model can also include an audit and playback module, which monitors KPI trends during the execution of defect detection formulation versions. These KPI trends include: acceptance consistency trends, maintenance cost trends, and formulation defect detection coverage trends that change over time. By setting up the audit and playback module, the KPI trends of the currently executed defect detection formulation version can be monitored during the execution process. This allows for real-time understanding of the changing trends in acceptance consistency, maintenance costs, and formulation defect detection coverage, enabling KPI trend tracking, auditing, and playback of defect detection performance. This achieves full-cycle effect tracking of defect detection formulation versions, facilitating optimization. For example, an industrial-grade verification barrier module can be used to suppress acceptance inconsistencies and untraceable outputs, ensuring the acceptance and delivery of defect detection processes for the tested devices. For example, the audit and playback module can support online inference and offline playback consistency verification, providing audit logs and change descriptions.
[0041] For example, refer to Figure 5 and Figure 10The defect detection management model can also include a scoring module and a rollback module. The scoring module is used to score the currently executed defect detection recipe version based on KPI trends during the execution of the defect detection recipe version. The rollback point execution rules can include: if the score of the currently executed defect detection recipe version is lower than the score of a historically stored defect detection recipe version in the recipe version record, the rollback module will roll back the defect detection recipe version to the historically stored defect detection recipe version based on the recipe version record, so that the score of the currently executed defect detection recipe version is the highest score in the recipe version record. For example, by monitoring the drift of defect detection results and KPIs, a defect detection recipe version update (change) can be triggered, and after replay evaluation, a canary release can be triggered to trigger the defect detection recipe execution.
[0042] For details, please refer to Figure 12 During the execution of a defect detection recipe version, the scoring module scores the currently executed defect detection recipe version in real time based on KPI trends. It then determines whether the score of the currently executed defect detection recipe version is lower than the score of a historically stored defect detection recipe version in the recipe version record. If the result is that the score of the currently executed defect detection recipe version is lower than the score of a historically stored defect detection recipe version, the rollback module is triggered to roll back the defect detection recipe version to a historically stored defect detection recipe version with a higher score than the currently executed defect detection recipe version. This ensures that the score of the currently executed defect detection recipe version is the highest in the recipe version record, guaranteeing the defect detection effect of the automatic defect detection equipment and enabling reusable defect detection recipe versions and closed-loop updates.
[0043] For example, refer to Figure 11 The statistics and calibration module can output tabular reports, field-based explanations, threshold change records, and export packages of defect detection execution recipe versions. The output includes result tables, evidence packages, defect detection execution recipe versions and version numbers, and supports playback, canary releases, and rollbacks. The field-based explanation output must reference at least one evidence field and its source.
[0044] There are several ways to score the currently executed defect detection recipe version based on KPI trends in the scoring module. Some of these methods are illustrated below.
[0045] For example, refer to Figure 10The scoring module obtains the score for the currently executed defect detection execution recipe version based on KPI trends. This can include: obtaining an acceptance consistency score based on acceptance consistency trends; obtaining a maintenance cost score based on maintenance cost trends; obtaining a recipe defect detection coverage score based on recipe defect detection coverage trends; and obtaining a weighted score for the currently executed defect detection execution recipe version based on the acceptance consistency score, maintenance cost score, and recipe defect detection coverage score.
[0046] For details, please refer to Figure 10 The scoring module determines the acceptance consistency score, maintenance cost score, and formula defect detection coverage score based on the acceptance consistency trend, maintenance cost trend, and formula defect detection coverage trend, respectively. Then, the acceptance consistency score, maintenance cost score, and formula defect detection coverage score are weighted and averaged to obtain the score of the currently executed defect detection execution formula version, thereby improving the accuracy of the determined score of the currently executed defect detection execution formula version.
[0047] It should be noted that the reference Figure 10 In the Acceptance Consistency Trend, a better Acceptance Consistency Trend indicates higher defect detection consistency, resulting in a higher Acceptance Consistency Score. In the Maintenance Cost Trend, higher Maintenance Costs indicate higher maintenance costs for the currently executed defect detection formulation version, resulting in a lower Maintenance Cost Score. In the Formulation Defect Detection Coverage Trend, higher Formulation Defect Detection Coverage indicates stronger defect detection coverage, resulting in a higher Formulation Defect Detection Coverage Score. Therefore, a higher score for the defect detection formulation version indicates a greater overall advantage in terms of Acceptance Consistency, Maintenance Costs, and Formulation Defect Detection Coverage.
[0048] The scoring module calculates the score of the currently executed defect detection execution recipe version by weighting the acceptance consistency score, maintenance cost score, and recipe defect detection coverage score. There are several ways to do this, and some examples are introduced below.
[0049] For example, refer to Figure 10The scoring module calculates a weighted score for the currently executed defect-detection recipe version based on the acceptance conformance score, maintenance cost score, and recipe defect detection coverage score. This can be achieved by: generating acceptance conformance weights, maintenance cost weights, and defect detection coverage weights based on the acceptance conformance index, maintenance cost index, and defect detection coverage index from the structured constraint indicators; multiplying the acceptance conformance score by the acceptance conformance weight to obtain a first product; multiplying the maintenance cost score by the maintenance cost weight to obtain a second product; and multiplying the recipe defect detection coverage score by the defect detection coverage weight to obtain a third product. The sum of the first, second, and third products is then used as the score for the currently executed defect-detection recipe version.
[0050] For details, please refer to Figure 10 The scoring module assigns weights to acceptance consistency, maintenance cost, and defect detection coverage metrics based on acceptance requirements within the structured constraint metrics. During weight assignment, if the acceptance consistency requirement is higher than the maintenance cost and defect detection coverage requirements, the acceptance consistency weight is assigned a larger value than the maintenance cost and defect detection coverage weights. Conversely, if the maintenance cost requirement is higher than the acceptance consistency and defect detection coverage requirements, the maintenance cost weight is assigned a larger value than the acceptance consistency and defect detection coverage weights. And if the defect detection coverage requirement is higher than the acceptance consistency and maintenance cost requirements, the defect detection coverage weight is assigned a larger value than both the acceptance consistency and maintenance cost weights. This allows for flexible adjustment of the acceptance consistency, maintenance cost, and defect detection coverage weights according to acceptance requirements, ensuring that the calculated defect detection execution formula version score better matches the acceptance requirements and that the optimization of the defect detection execution formula version more closely approximates or even meets the acceptance requirements.
[0051] It should be noted that the above-mentioned functional modules can be configured through hardware such as memory and processor, as well as software code running on them.
[0052] In summary, for reference Figure 13The defect detection management model of the automatic optical inspection equipment provided in this disclosure has the following beneficial effects: (1) The defect detection management model of the automatic optical inspection equipment provided in this disclosure can maintain a stable judgment caliber under process / structure fluctuation environment, and improve generalization and robustness. (2) The defect detection management model outputs auditable evidence and tabular semantic explanation, which improves interpretability. (3) The defect detection execution recipe version is reusable, replayable, and versionable, which reduces maintenance costs. (4) Executable execution rules and defect detection execution recipe versions (DSL / configuration) are automatically generated and bound to defect detection execution recipe version management, gray release, and rollback point.
[0053] It should be noted that, in addition to the structure described above, the defect detection management model of the automatic optical inspection equipment in this embodiment may also include the related structures described in part of the embodiment, all of which are within the protection scope of the defect detection management model of the automatic optical inspection equipment provided in this embodiment.
[0054] Example 2 This disclosure provides an automated optical inspection device, which includes a defect detection management model for any of the automated optical inspection devices provided in the first aspect of this disclosure.
[0055] Although embodiments of the present disclosure have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present disclosure, and such modifications and variations all fall within the scope defined by the appended claims.
Claims
1. A defect detection management model for an automated optical inspection device, characterized in that, include: The demand access module is used to receive acceptance requirements for the device under test. The constraint parsing module is used to perform structured parsing of the acceptance requirements to obtain at least one structured constraint indicator. A threshold-related parameter generation module is used to generate the threshold parameters for each of the structured constraint indicators; The rule writing module is used to generate defect detection execution rules based on the structured constraint indicators and the indicator threshold parameters of each structured constraint indicator; The statistics and calibration module is used to adjust the index threshold parameters based on the actual defects of the device under test and the detected defects obtained by executing the defect detection execution rules, and to update the defect detection execution rules based on the adjusted index threshold parameters.
2. The defect detection management model as described in claim 1, characterized in that, The structured constraint indicators include at least some of the following indicators: detection area indicators, defect type indicators, false detection risk budget indicators, acceptance standard indicators, allowable threshold range indicators, exception situation indicators, defect detection execution action indicators, defect interpretation indicators, acceptance consistency indicators, maintenance cost indicators, and defect detection coverage indicators.
3. The defect detection management model as described in claim 1, characterized in that, The rule writing module determines the defect detection execution rule for each structured constraint indicator and its corresponding indicator threshold parameter, based on the large language model, specific domain language model, rule template library and parameter binding rules corresponding to the structured constraint indicator. The defect detection management model also includes a recipe encapsulation module, which is used to encapsulate the defect detection execution rules of all the structured constraint indicators based on versioned encapsulation rules to obtain the defect detection execution recipe version. The defect detection execution recipe version also includes syntax verification execution rules, range verification execution rules, evidence field constraint execution rules, gray release execution rules, and rollback point execution rules.
4. The defect detection management model as described in claim 3, characterized in that, The statistics and calibration module includes: The actual defect statistics module is used to collect actual defect data of the tested device; The defect detection statistics module is used to statistically analyze the defect detection data obtained by executing the defect detection execution rules. The verification guardrail module is used to calculate the actual false detection data from the actual defect data and the detected defect data, and to determine whether the actual false detection data meets the false detection risk budget index in the structured constraint index. The quantile mapping module is used to process the actual defect data and the detected defect data to obtain the quantile mapping result; The threshold adaptive module is used to generate suggested threshold parameters for the index based on the actual false detection data and the quantile mapping result when the actual false detection data does not meet the false detection risk budget index in the structured constraint index.
5. The defect detection management model as described in claim 4, characterized in that, The threshold adaptive module is also used for: When the actual false detection data does not meet the false detection risk budget index in the structured constraint index, a change description and a change impact assessment are generated based on the actual false detection data and the quantile mapping result.
6. The defect detection management model as described in claim 4, characterized in that, The defect detection management model further includes: a version and canary release module, which is used to update the suggested parameter of the indicator threshold to the indicator threshold parameter based on the canary release execution rule; The threshold adaptive module is further configured to: after updating the indicator threshold suggestion parameter to the indicator threshold parameter, store the defect detection execution recipe version before the change of the indicator threshold parameter and the defect detection execution recipe version after the change to obtain a recipe version record, which can be used when rolling back according to the rollback point execution rule in the defect detection execution rule.
7. The defect detection management model as described in claim 6, characterized in that, Also includes: An audit and replay module is used to monitor KPI trends when executing the defect detection recipe version; The KPI trends include: acceptance consistency trend, maintenance cost trend, and formula defect detection coverage trend, which change over time as the defect detection is performed on the formula version.
8. The defect detection management model as described in claim 7, characterized in that, Also includes: The scoring module is used to score the currently executed defect detection execution recipe version based on the KPI trend during the execution of the defect detection execution recipe version; The rollback module, wherein the rollback point execution rules include: when the score of the currently executed defect detection execution recipe version is less than the score of the historically stored defect detection execution recipe version in the recipe version record, the rollback module rolls back the defect detection execution recipe version to the historically stored defect detection execution recipe version based on the recipe version record, so that the score of the currently executed defect detection execution recipe version is the highest-scoring defect detection execution recipe version in the recipe version record.
9. The defect detection management model as described in claim 8, characterized in that, The scoring module scores the currently executed defect detection execution recipe version based on the KPI trend, including: The scoring module obtains an acceptance consistency score based on the acceptance consistency trend. The scoring module generates a maintenance cost score based on the maintenance cost trend. The scoring module obtains a formula defect detection coverage score based on the trend of the formula defect detection coverage. The scoring module calculates the score of the currently executed defect detection execution recipe version by weighting the acceptance consistency score, the maintenance cost score, and the recipe defect detection coverage score.
10. The defect detection management model as described in claim 9, characterized in that, The scoring module calculates a weighted score for the currently executed defect detection execution recipe version based on the acceptance consistency score, the maintenance cost score, and the recipe defect detection coverage score, including: The scoring module generates acceptance consistency weight, maintenance cost weight, and defect detection coverage weight based on the acceptance consistency index, maintenance cost index, and defect detection coverage index in the structured constraint indicators, respectively. The scoring module multiplies the acceptance consistency score by the acceptance consistency weight to obtain a first product; The scoring module multiplies the maintenance cost score by the maintenance cost weight to obtain a second product; The scoring module multiplies the formula defect detection coverage score by the defect detection coverage weight to obtain a third product; The scoring module uses the sum of the first product, the second product, and the third product as the score for the currently executed defect detection execution recipe version.